tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-28 04:15:21 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama : deprecate llama_kv_self_ API (#14030)

Browse Source 
* llama : deprecate llama_kv_self_ API

ggml-ci

* llama : allow llama_memory_(nullptr)

ggml-ci

* memory : add flag for optional data clear in llama_memory_clear

ggml-ci
This commit is contained in:
Georgi Gerganov
2025-06-06 14:11:15 +03:00
committed by GitHub
parent 487a5e0401
commit 745aa5319b
34 changed files with 206 additions and 127 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
tools/main/main.cpp
View File

@ -147,6 +147,8 @@ int main(int argc, char ** argv) {
return 1;
}
auto * mem = llama_get_memory(ctx);
const llama_vocab * vocab = llama_model_get_vocab(model);
auto chat_templates = common_chat_templates_init(model, params.chat_template);
@ -351,7 +353,7 @@ int main(int argc, char ** argv) {
}
// remove any "future" tokens that we might have inherited from the previous session
llama_kv_self_seq_rm(ctx, -1, n_matching_session_tokens, -1);
llama_memory_seq_rm(mem, -1, n_matching_session_tokens, -1);
}
LOG_DBG("recalculate the cached logits (check): embd_inp.size() %zu, n_matching_session_tokens %zu, embd_inp.size() %zu, session_tokens.size() %zu\n",
@ -599,8 +601,8 @@ int main(int argc, char ** argv) {
LOG_DBG("context full, swapping: n_past = %d, n_left = %d, n_ctx = %d, n_keep = %d, n_discard = %d\n",
n_past, n_left, n_ctx, params.n_keep, n_discard);
llama_kv_self_seq_rm (ctx, 0, params.n_keep , params.n_keep + n_discard);
llama_kv_self_seq_add(ctx, 0, params.n_keep + n_discard, n_past, -n_discard);
llama_memory_seq_rm (mem, 0, params.n_keep , params.n_keep + n_discard);
llama_memory_seq_add(mem, 0, params.n_keep + n_discard, n_past, -n_discard);
n_past -= n_discard;
@ -623,9 +625,9 @@ int main(int argc, char ** argv) {
LOG_DBG("div: [%6d, %6d] / %6d -> [%6d, %6d]\n", ga_i + ib*bd, ga_i + ib*bd + ga_w, ga_n, (ga_i + ib*bd)/ga_n, (ga_i + ib*bd + ga_w)/ga_n);
LOG_DBG("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", ga_i + ib*bd + ga_w, n_past + ib*bd, dd, ga_i + ib*bd + ga_w + dd, n_past + ib*bd + dd);
llama_kv_self_seq_add(ctx, 0, ga_i, n_past, ib*bd);
llama_kv_self_seq_div(ctx, 0, ga_i + ib*bd, ga_i + ib*bd + ga_w, ga_n);
llama_kv_self_seq_add(ctx, 0, ga_i + ib*bd + ga_w, n_past + ib*bd, dd);
llama_memory_seq_add(mem, 0, ga_i, n_past, ib*bd);
llama_memory_seq_div(mem, 0, ga_i + ib*bd, ga_i + ib*bd + ga_w, ga_n);
llama_memory_seq_add(mem, 0, ga_i + ib*bd + ga_w, n_past + ib*bd, dd);
n_past -= bd;